Feature or enhancement
Proposal:
Very similar to #148729. I noticed that for regexes of the form
re.compile("^foo", re.MULTILINE)
there's still a character by character loop calling SRE(match).
The SRE(match) function call is expensive and can be avoided by scanning for newline characters first. The newline character scan can be vectorized in the UCS1 case by calling memchr for additional speedup.
A patch would look like this:
diff --git a/Modules/_sre/sre_lib.h b/Modules/_sre/sre_lib.h
index 4f1269988b9..11d341ea2ab 100644
--- a/Modules/_sre/sre_lib.h
+++ b/Modules/_sre/sre_lib.h
@@ -1869,12 +1869,42 @@ SRE(search)(SRE_STATE* state, SRE_CODE* pattern)
state->start = state->ptr = ptr = end;
return 0;
}
- while (status == 0 && ptr < end) {
- ptr++;
- RESET_CAPTURE_GROUP();
- TRACE(("|%p|%p|SEARCH\n", pattern, ptr));
- state->start = state->ptr = ptr;
- status = SRE(match)(state, pattern, 0);
+ if (pattern[0] == SRE_OP_AT &&
+ pattern[1] == SRE_AT_BEGINNING_LINE)
+ {
+ /* Skip to line boundary */
+ while (status == 0 && ptr < end) {
+ ptr++;
+ if (!SRE_IS_LINEBREAK((int) ptr[-1])) {
+#if SIZEOF_SRE_CHAR == 1
+ ptr = (SRE_CHAR *)memchr(ptr, '\n', end - ptr);
+ if (!ptr) {
+ break;
+ }
+#else
+ while (ptr < end && !SRE_IS_LINEBREAK((int) *ptr)) {
+ ptr++;
+ }
+ if (ptr >= end) {
+ break;
+ }
+#endif
+ ptr++;
+ }
+ RESET_CAPTURE_GROUP();
+ TRACE(("|%p|%p|SEARCH\n", pattern, ptr));
+ state->start = state->ptr = ptr;
+ status = SRE(match)(state, pattern, 0);
+ }
+ }
+ else {
+ while (status == 0 && ptr < end) {
+ ptr++;
+ RESET_CAPTURE_GROUP();
+ TRACE(("|%p|%p|SEARCH\n", pattern, ptr));
+ state->start = state->ptr = ptr;
+ status = SRE(match)(state, pattern, 0);
+ }
}
}
Has this already been discussed elsewhere?
This is a minor feature, which does not need previous discussion elsewhere
Links to previous discussion of this feature:
No response
Linked PRs
Feature or enhancement
Proposal:
Very similar to #148729. I noticed that for regexes of the form
there's still a character by character loop calling
SRE(match).The
SRE(match)function call is expensive and can be avoided by scanning for newline characters first. The newline character scan can be vectorized in the UCS1 case by callingmemchrfor additional speedup.A patch would look like this:
Has this already been discussed elsewhere?
This is a minor feature, which does not need previous discussion elsewhere
Links to previous discussion of this feature:
No response
Linked PRs